How Virtual Machinery Can Bridge The Explanatory Gap In Natural and - - PowerPoint PPT Presentation

Seminar in School of Biosciences, University of Birmingham 11 May 2010 Talk at SAB2010, Clos-Luc` e, Amboise, France, 28th Aug 2010

How Virtual Machinery Can Bridge The “Explanatory Gap”

In Natural and Artificial Systems.

Presented at SAB2010: http://www.sab2010.org/ Aaron Sloman

School of Computer Science, University of Birmingham http://www.cs.bham.ac.uk/∼axs/

These slides are available in my ‘talks’ directory: http://www.cs.bham.ac.uk/research/projects/cogaff/talks/#sab2010 The paper presenting these ideas (for SAB2010) is online here: http://www.cs.bham.ac.uk/research/projects/cogaff/10.html#sab Videos of presentations at SAB2010 are at http://www.cs.bham.ac.uk/research/projects/cogaff/movies/#sabtalks including this presentation: http://rtsp://video.cpm.upmc.fr/videocpm/2010/sab 2010/aaron sloman.rm

Biosciences May 2010 & SAB2010 Slide 1 Last revised: January 13, 2011

Alternative titles

• 1. How could evolution

(or anything else) get ghosts into machines?

• 2. Helping Darwin:

How to Think About Evolution of Consciousness

Biosciences May 2010 & SAB2010 Slide 2 Last revised: January 13, 2011

Apology

My slides are too cluttered to read during a presentation. I’ll select bits to talk about today, and put the whole thing on my talks web site later.

http://www.cs.bham.ac.uk/research/projects/cogaff/talks/#sab2010 NOTE: These slides include a topic not discussed in the published paper, namely the importance of understanding evolutionary and developmental transitions in information-processing systems. This is an extension of the discussion of transitions in evolution by John Maynard Smith and E¨

• rs

Szathm´ ary (1995).

Biosciences May 2010 & SAB2010 Slide 3 Last revised: January 13, 2011

Abstract of the Paper (Part 1)

Many of Darwin’s opponents, and some of those who accepted the theory of evolution as regards physical forms, objected to the claim that human mental functions, and consciousness in particular, could be products of evolution. There were several reasons for this opposition, including unanswered questions as to how physical mechanisms could produce mental states and processes – an old, and still surviving, philosophical problem. A new answer is now available. Evolution could have produced the “mysterious” aspects

• f consciousness if, like engineers developing computing systems in the last six or seven

decades, evolution encountered and “solved” increasingly complex problems of representation and control (including self-monitoring and self-control) by using systems with increasingly abstract mechanisms based on virtual machines, including most recently self-monitoring virtual machines. These capabilities are, like many capabilities of computer-based systems, implemented in non-physical virtual machinery which, in turn, are implemented in lower level physical mechanisms. This would require far more complex virtual machines than human engineers have so far

• created. No-one knows whether the biological virtual machines could have been

implemented in the discrete-switch technology used in current computers. These ideas were not available to Darwin and his contemporaries: most of the concepts, and the technology, involved in creation and use of sophisticated virtual machines were developed only in the last half century, as a by-product of a large number of design decisions by hardware and software engineers solving different problems.

Biosciences May 2010 & SAB2010 Slide 4 Last revised: January 13, 2011

Abstract (Part 2)

I shall try to characterise some of the forms of control and self-control that are made possible by the use of “Running Virtual Machines” (RVMs). I suggest that biological evolution “discovered” the need for them long before we did and has produced more complex and varied kinds of virtual machinery than we have so far. All this has implications for the future of machine intelligence (and more generally for development of increasingly robust, autonomous systems). It also has implications for the future of philosophy of mind, and looks likely to provide answers to challenges to Darwin’s theories from several of his competitors and supporters, including Thomas Huxley. It will finally show us how to bridge what has been called “the explanatory gap” between the physical and the mental, which was, and still is, a serious problem for Darwin’s theory

• f evolution.

For a more detailed introduction to the philosophical arguments related to this topic see

http://www.cs.bham.ac.uk/research/projects/cogaff/talks/#cons09

This is a trial run for a talk to be presented at SAB2010 http://www.sab2010.org/ The conference proceedings paper is here

http://www.cs.bham.ac.uk/research/projects/cogaff/10.html#sab

To illustrate virtual machinery I’ll start with a simple demo (using the Pop-11 SimAgent toolkit.)

http://www.cs.bham.ac.uk/research/projects/poplog/figs/simagent

Biosciences May 2010 & SAB2010 Slide 5 Last revised: January 13, 2011

Let’s vote

Do you agree with the following?

• Ignorance can cause poverty?
• Poverty can cause crime – e.g. stealing property?
• Over-zealous selling of mortgages can cause hardship for people in

many countries? Does anyone know why I am asking these questions? I’ll explain later. MEANWHILE: Think of a number - and remember what it was.

Biosciences May 2010 & SAB2010 Slide 6 Last revised: January 13, 2011

Ryle’s “Ghost in the machine”

In 1949 Gilbert Ryle’s book The Concept of Mind was published. In that book, he introduced the phrase “The Ghost in the Machine”.

Chapter 1 was about “Descartes’ Myth” according to which mental processes occur in an “occult stream of consciousness” to which the owner has “privileged access”, and about which others (onlookers, critics, teachers, biographers, and friends) can never be sure their beliefs or comments “have any vestige of truth”.

(page 15)

Ryle refers to this as “the Official Doctrine”, and says he will often refer to it “with deliberate abusiveness”, as “the dogma of the Ghost in the Machine”, which he hopes to prove “entirely false”.

Biosciences May 2010 & SAB2010 Slide 7 Last revised: January 13, 2011

We can move beyond Ryle

I have great admiration for Ryle’s work (and learnt from him as a student). But I hope to show that neither he, nor the defenders of “the official doctrine” understood the conditions under which a version of that doctrine can turn out to be entirely true, as a product of biological evolution.

Ryle had a good excuse: his book was written before most of the scientific and technological developments had occurred that make it possible for us now to put ghosts into millions of machines. He could not have known about: – device drivers, – memory management systems, – paged virtual memory, – garbage collection, – compilers, interpreters, incremental compilers, just-in-time compilers, – time-sharing operating systems, – network protocols, – file-system management, – firmware, ....

I shall try to explain the significance of those developments later.

What do all these things combine to produce?

Biosciences May 2010 & SAB2010 Slide 8 Last revised: January 13, 2011

We still don’t know how to put ghosts with animal intelligence into machines

But we can make some suggestions about where to look for answers. Every intelligent ghost must contain An information processing machine It’s a very long term research programme.

Biosciences May 2010 & SAB2010 Slide 9 Last revised: January 13, 2011

One of the key ideas we need is “Running Virtual Machine” (RVM)

• A machine is a complex system composed of sub-systems that interact

causally with one another and with things in the machine’s environment.

• A physical machine has only parts and behaviours that can be fully

described using the concepts of the physical sciences.

e.g. the concepts of physics and chemistry.

• Some machines, e.g. a chess-playing machine, will do things and have

capabilities whose description requires concepts not definable in terms

• f the concepts of physics and chemistry.
• We call such a thing a virtual machine or more precisely a Running

Virtual Machine (RVM).

• In order to exist and do anything, such a RVM will need to be

implemented in a physical machine.

For more on this see

http://www.cs.bham.ac.uk/research/projects/cogaff/talks/#mos09

Biosciences May 2010 & SAB2010 Slide 10 Last revised: January 13, 2011

Virtual machines are everywhere

Many produced by biological evolution Naturally occurring machines and RVMs are mainly on the left. Man-made machines and RVMs on the right.

How many levels of (relatively) virtual machinery does physics itself require?

Biosciences May 2010 & SAB2010 Slide 11 Last revised: January 13, 2011

Matter, Energy and Information

The portion of SAB2010 at which this was presented was held a museum of Leonardo da vinci’s work at Clos-Luc´ e, Amboise. On the bus from Paris we had a video presentation on Leonardo’s life and work. One of things I learnt was that he had written: “Everything is motion.” He seems to have been saying that the world can be seen as matter and energy interacting and constantly changing features and relationships. I suspect that if he were alive today, he would agree that Besides matter and energy there is information and Besides many varieties of motion there are many varieties of information-processing Our understanding of that variety is still in its early stages – only since the 20th century have we begun to understand how to make anything but the very simplest information processing machines. There were some precursors, e.g. Charles Babbage and Ada Lovelace – who anticipated some of Turing’s ideas about the significance of computers. An important feature of information processing is that it can be a form of control: information can specify what to do, how to do it, how to weigh up conflicting alternatives, and many more – These are major aspects of animal information processing.

Biosciences May 2010 & SAB2010 Slide 12 Last revised: January 13, 2011

Main points - I (See proceedings paper.)

• A problem for Darwin’s theory of evolution by natural selection
• Can his theory be applied to evolution of mind and consciousness?
• Even some of his supporters thought not.
• Wallace – the co-inventor of the theory thought not.
• There seemed to be a serious problem of the relation between mind

and matter.

Much discussed by Darwin’s contemporaries – friends and foes.

Biosciences May 2010 & SAB2010 Slide 13 Last revised: January 13, 2011

Main points - II

• Later work in philosophy, psychology, neuroscience, AI/Robotics, kept returning to this

problem, but it is still unresolved.

Often the problem is reformulated, e.g. – “the problem of qualia”, – “the problem of phenomenal consciousness”, – “the hard problem of consciousness”, ... These are all essentially the same problem, though it is multi-faceted.

• I offer an explanation for the difficulty and a research programme to develop a

solution, based on the concept of a Running Virtual Machine.

• We can give scientific and engineering respectability to notions of emergence and

downward causation, using ideas from computing systems.

• This relates closely to one of the oldest unsolved philosophical problems:

What do we mean by X caused Y?

Biosciences May 2010 & SAB2010 Slide 14 Last revised: January 13, 2011

What do we mean by X caused Y?

Is causation mere correlation (David Hume)

• r something deeper (Immanuel Kant)?

Can events in virtual machines cause physical events? How?

You implied they can if you answered the questions about ignorance, poverty and crime, positively.

Can mental events cause physical events? How?

If I ask people to put up their left hand if they chose a positive integer, or their right hand if they a negative integer, or both if they chose a number that is not an integer, then their physical action will be caused in part by the mental process of selecting a number. If relative poverty causes crime, and some of the crimes involve people being robbed or stabbed, then social virtual machinery can cause physical events and processes to occur.

Biosciences May 2010 & SAB2010 Slide 15 Last revised: January 13, 2011

A framework for what I am doing

John Maynard Smith and E¨

• rs Szathm´

ary (1995) proposed that there are eight major transitions in evolution, summarised here:

http://en.wikipedia.org/wiki/The Major Transitions in Evolution

Transition from: Transition to: 1 Replicating molecules “Populations” of molecules in compartments 2 Independent replicators (probably RNA) Chromosomes 3 RNA as both genes and enzymes DNA as genes; proteins as enzymes 4 Prokaryotes Eukaryotes 5 Asexual clones Sexual populations 6 Protists Multicellular organisms - animals, plants, fungi 7 Solitary individuals Colonies with non-reproductive castes 8 Primate societies Human societies with language, enabling memes

Biosciences May 2010 & SAB2010 Slide 16 Last revised: January 13, 2011

They identified features common to the eight transitions:

• 1. Smaller entities come together to form larger entities.
• 2. Smaller entities become differentiated as part of a larger entity.
• 3. Smaller entities become unable to replicate in the absence of the larger

entity.

• 4. Smaller entities become able to disrupt the development of a larger

entity.

• 5. New ways arise of transmitting information.

Is that the only kind of information-processing transition? Suggestion: We should try to produce a taxonomy of types of evolutionary

• r developmental transition (gradual or discontinuous) connected with

information processing – in animals, and in future animats.

(A century or two should suffice.)

Biosciences May 2010 & SAB2010 Slide 17 Last revised: January 13, 2011

Another set of transitions

We can generalise the last feature to include

• new kinds of information
• new ways of acquiring, processing, or using information.
• new information-processing architectures

These changes can be related to different sorts of trajectories in evolution and in development. Trajectories in different spaces

• Trajectories in the space of possible sets of requirements (niches)
• Trajectories in the space of designs for systems satisfying requirements
• Trajectories in The space of implementations for each design

Biosciences May 2010 & SAB2010 Slide 18 Last revised: January 13, 2011

Spaces for abstract dynamics

Transitions (and interactions between transitions) can occur

• within the space of possible designs
• within the space of possible sets of requirements (niches)

Both designs and requirements can exist without a designer or a requirer.

Biosciences May 2010 & SAB2010 Slide 19 Last revised: January 13, 2011

Studying requirements

I have learnt over the last 40 years or so that understanding the requirements that organisms evolve or develop to meet can be very hard. Many researchers assume the requirements are clear and the only problem is to find designs and implementations. Example: assuming vision is mainly about recognition. It isn’t. (A topic for another occasion) This is an important part of AIIB: AI-Inspired Biology

Biosciences May 2010 & SAB2010 Slide 20 Last revised: January 13, 2011

The “Explanatory Gap”

A problem that puzzled Darwin and fired up his critics:

• There’s lots of evidence for evolution of physical forms.
• There’s no similar evidence that human minds could be products of evolution.
• There seems to be no way that physical matter can produce mental processes.

This is the so-called “explanatory gap”

Biosciences May 2010 & SAB2010 Slide 21 Last revised: January 13, 2011

The “Explanatory Gap”

A problem that puzzled Darwin and fired up his critics:

• There’s lots of evidence for evolution of physical forms.
• There’s no similar evidence that human minds could be products of evolution.
• There seems to be no way that physical matter can produce mental processes.

This is the so-called “explanatory gap”

• Until the last few decades, explanatory mechanisms linking physical and mental

phenomena were not even conceivable to most scientists: hence the “explanatory gap” of Huxley and others and Chalmers’ “Hard problem of consciousness”, etc.

• Now, as a result of a great deal of work on hardware, software, firmware, and CS

theory, we know how to make things that have some of the features required for working explanatory models of mental processes, with some of the key features of mental processes (including having causal powers, without being physical processes) – but only in very simplified form.

• Most philosophers, psychologists and neuroscientists have ignored or misunderstood

this, and so have many AI/Computing/Software researchers.

I’ll give some pointers, but not explain in detail, below.

• There are deep implications for philosophical analyses of causation.

E.g. downward causation from mind-like events and processes to physical events and processes.

Biosciences May 2010 & SAB2010 Slide 22 Last revised: January 13, 2011

Some quotes

Several thinkers in Darwin’s time mentioned the problem.

T.H. Huxley

“How it is that anything so remarkable as a state of consciousness comes about as a result of irritating nervous tissue, is just as unaccountable as the appearance of the Djin when Aladdin rubbed his lamp”. Lessons in Elementary Physiology (1866) (Where exactly?)

• G. J. Romanes

“We know by immediate or subjective analysis that consciousness only occurs when a nerve-centre is engaged in such a focusing of vivid or comparatively unusual stimuli as have been described; and when as a preliminary to this focusing or act of discriminative adjustment there arises in the nerve-centre a comparative turmoil of stimuli coursing in more or less unaccustomed directions, and therefore giving rise to a comparative delay in the occurrence of the eventual response. But we are totally in the dark as to the causal connection, if any, between such a state of turmoil in the ganglion and the occurrence of consciousness.”(p75) Mental evolution in animals (1883) (Quoted by Whittaker in his review in Mind, 1884) There were many others, both before and after Darwin’s time, and searching through journals published after Origin of Species there is much to be rediscovered about views of the mind/matter problem in the 19th century. It is much easier to do now that so many old journal issues have been digitised, e.g. Mind. I’ll start by helping you understand, at first hand, what all the fuss was about.

Biosciences May 2010 & SAB2010 Slide 23 Last revised: January 13, 2011

Do some “consciousness science”

Ontologies for conscious contents

Stare at each of these two pictures for a while. Each is ambiguous and should flip between (at least) two very different views. Try to describe exactly what changes when the flip occurs. What concepts are needed for the different experiences?

In one case geometrical relations and distances change. In the other case geometry is unchanged, but biological functions change. Can a cube be experienced as “looking to left or to right”? If not, why not?

Nothing changes on the screen or in the optical information entering your eyes and brain. Compare the kind of vocabulary used to describe parts and relationships in the two views

• f the Necker cube, and in the two views of the “duck-rabbit”.

So contents of consciousness can be three dimensional and can include information about entities with functional roles (e.g. eye, mouth, bill), mental states (e.g. looking left), ...

Biosciences May 2010 & SAB2010 Slide 24 Last revised: January 13, 2011

Empirical basis for referring to contents of consciousness

Introspection provides empirical evidence for mental contents “inside” us, distinct from external causes or internal physical processes:

• Ambiguous figures: as you stare at them, nothing changes out there, only the

experiences/qualia (in your mind) “flip” (E.g. Necker cube, face-vase, old/young lady, etc.)

This one rotates in 3-D in either direction: http://www.procreo.jp/labo/labo13.html

• Optical illusions (of many kinds): Muller-Lyer, Ebbinghaus, motion/colour after-effects.
• Dreams, hallucinations, hypnotic suggestions, effects of alcohol and other drugs.
• Different people see the same things differently. E.g. short-sighted and long-sighted people.

Identical twins look different to people who know them well, but look the same to others.

• Cf. Botanical expertise makes similar plants look different. Colour blindness.
• Pressing eyeball makes things appear to move when they don’t, and can undo binocular fusion: you get

two percepts of the same object; crossing your eyes can also do that.

• Put one hand into a pail of hot water, the other into a pail of cold water, then put both into lukewarm

water: it will feel cool to one hand and warm to the other. (A very old philosophical experiment.)

• People and other things look tiny from a great height – without any real change in size.
• Aspect ratios, what’s visible, specularities, optical flow – all change with viewpoint.
• We experience only portions of things. A cube has six faces but we can’t see them all: what you

experience changes as you move.

• Thinking, planning, reminiscing, daydreaming, imagining, can be done with eyes closed ....
• Composing poems, or music, or proving theorems with your eyes shut.

Rich mental contents (not perceptual contents) can be involved in all of these.

Biosciences May 2010 & SAB2010 Slide 25 Last revised: January 13, 2011

We tend not to notice the diversity of contents

The adjectival phrase “conscious of” is less deceptive than the noun “consciousness”. We can be conscious of many very different things, with different implications (the concept is polymorphic):

E.g. there are great differences between being conscious of

• something moving towards you
• something looking at you
• a door being opened
• being close to a cliff edge
• being half asleep
• being horizontal
• being unpopular
• being 25 years old
• being in France
• being able to cook a meal without help
• being unknown to anyone in the room
• being deaf
• being more knowledgeable than most of the population
• being on the verge of a great discovery

X being conscious of Y involves X, or some part of X, having access to information about Y – but what that amounts to differs according to what X is and what Y is.

Biosciences May 2010 & SAB2010 Slide 26 Last revised: January 13, 2011

Responses to the problem – up to now

The variations include:

• rejection of the problem as somehow due to a deep muddle
• claiming that it is a real problem but lacking any solution that human minds can

understand,

• offering a reformulation of the problem alleged to solve it
• resurrecting the problem with a new label
• proposing a philosophical or scientific research project to solve it
• offering specific solutions that appeal to recent advances in physics or mathematics
• assembling experimental and observational data about it
• producing working computer models of various kinds, e.g.
• developing new technical philosophical concepts in the hope of clarifying it

(e.g. supervenience)

• and many more.

To understand how minds evolved, we need to understand what they do – i.e. what their functions are – and also their diversity.

Biosciences May 2010 & SAB2010 Slide 27 Last revised: January 13, 2011

Opinions (in Darwin’s time and now) differed as to

• Whether there is

– one divide (a dichotomy) between things with and things without consciousness

• r

– many discontinuities in mental competences

• Whether it is all a matter of degree (e.g. Susan Greenfield?)

i.e. continuous variation from most simple to most complex. NB: Biological changes cannot be continuous, if (a) everything is implemented in chemical structures (b) there is no inheritance of acquired characteristics: Between any two times there can be only a finite number of generations: ruling out continuous change, though many small discontinuities can occur. (It is not always noticed that biology and continuity are incompatible.)

• Whether the varieties of mind can be arranged in some sort of linear order

NO according to Whittaker, reviewing Romanes in Mind 1884.

• Whether it is possible for matter to produce mind

(a problem that has many different formulations)

• Whether it is possible for mind to influence matter

If minds were produced by biological evolution the answer must be YES. Minds are information-based control systems, required by ALL organisms.

Biosciences May 2010 & SAB2010 Slide 28 Last revised: January 13, 2011

Figure for Neural Correlates of Consciousness

by Mormann & Koch (From Scholarpedia)

The figure summarises a widely-held, but mistaken view of consciousness.

http://www.scholarpedia.org/article/Neural correlates of consciousness

Biosciences May 2010 & SAB2010 Slide 29 Last revised: January 13, 2011

A common mistake about consciousness

The mistake is to assume that causation goes only one way.

We also need to understand how perceptual experiences, pleasures, pains, desires, motives, decisions, intentions, plans, preferences.... can produce behaviour? Epiphenomenalism says they cannot: they are caused but cannot be causes. BUT Mental phenomena are products of biological evolution. They have a crucial biological role in control: of actions, of energy deployment, of goal-selection, of plan-formation, of plan execution, of problem solving. SO A theory of mental phenomena must not ignore causal powers of mental events and processes, including

• their ability to cause other mental phenomena
• and their ability to cause physical behaviour

That means we have to explain how “downward causation” is possible. We have learnt a lot about this in the last 50-60 years, but it has mostly gone unnoticed.

Biosciences May 2010 & SAB2010 Slide 30 Last revised: January 13, 2011

Transitions in information processing

I want to show how some of the transitions in what we have designed and built in complex computing systems can suggest a strategy for helping Darwin answer his critics who think evolution could not have produced mental processes. Transitions in requirements Transitions in designs and implementations

(Including development of new mechanisms for use in implementations.)

Especially the use of running virtual machines of various kinds

Biosciences May 2010 & SAB2010 Slide 31 Last revised: January 13, 2011

Virtual machines are everywhere

Many produced by biological evolution

How many levels of (relatively) virtual machinery does physics itself require?

Biosciences May 2010 & SAB2010 Slide 32 Last revised: January 13, 2011

Why the vote?

Earlier, I asked: Do you agree with the following?

• Ignorance can cause poverty?
• Poverty can cause crime – e.g. stealing property?
• Over-zealous selling of mortgages can cause hardship for people in

many countries?

If you think such causes CAN produce effects, then it appears that you have social and/or socio-economic virtual machines in your ontology. (That’s true of most people in our culture.) Moreover you think virtual machine states and processes can cause physical events (like television sets moving through broken windows when poverty causes crime). What having virtual machines in your ontology means is another (hard) question, partially answered later.

Biosciences May 2010 & SAB2010 Slide 33 Last revised: January 13, 2011

The “gap” closing

The last half century’s advances in development of running virtual machines at last provide a basis for closing what has been called “the explanatory gap”.

(But only if the right notion of virtual machine is used: a concept many people understand only dimly.)

• This gap was identified as a real problem for Darwin’s theory of evolution, even among

people who were convinced by evidence for evolution of physical forms

(e.g. T.H. Huxley, though the physicist Tyndall had earlier discussed it at as a mystery)

• The problem for Darwinians was that there was plenty of evidence for gradual

evolution of physical forms between various species, including humans, but the transition from non-human to human minds seemed to involve such a big discontinuity that there was no comparable evidence.

• Further, some people thought it was inconceivable that an explanation of how physical

bodies could produce minds would ever be forthcoming.

• Only now are we on the threshold of understanding what evolution might have had to

do.

• But we need to deal with a bi-directional gap: we need to understand both

– how physical mechanisms, and their states and processes, produce and support mental processes, – how mental states and processes have control functions, i.e. produce, monitor and modulate physical processes: as virtual machinery does in many computing systems.

Biosciences May 2010 & SAB2010 Slide 34 Last revised: January 13, 2011

KEY IDEA: Running Virtual Machine (RVM)

The idea of a running virtual machine (RVM) should not be confused with the abstract mathematical structure defining a type of VM, which can be thought of as a “Mathematical Model” (MM), about which theorems can be proved, etc., but which does not do anything, anymore than numbers do. Distinguish: PMs MMs RVMs Illustrate with demos.

E.g. Sheepdog demo. See Movie 5 here:

http://www.cs.bham.ac.uk/research/projects/poplog/figs/simagent

Biosciences May 2010 & SAB2010 Slide 35 Last revised: January 13, 2011

Biological conjecture

I conjecture that biological evolution encountered those design problems long before we did and produced solutions using virtual machinery long before we did – in order to enable organisms to deal with rapidly changing and complex information structures,

e.g. in visual perception, decision making, control of actions, self-monitoring etc.

• You can’t rapidly rewire millions of neurons when you look in a new direction
• or when you switch from approaching prey to deciding in which direction to try to

escape from a new predator, using visible details of the terrain.

• So there’s no alternative to using virtual machinery.
• But we know very little about biological virtual machinery.

Nobody knows how brain mechanisms provide virtual machinery that supports proving geometric theorems, thinking about infinite sets of numbers, or algebra, or wanting to rule the world.

• The visual competences demonstrated here remain unexplained

http://www.cs.bham.ac.uk/research/projects/cogaff/misc/multipic-challenge.pdf

We know that humans can use very different languages to say and do similar things (e.g. teach physics, discuss the weather); but evolution could not have produced special unique brain mechanisms for each language (since most are too new) – it’s more likely that language learning creates specialised VMs running on more general physiological mechanisms. Some conjectures about evolution of language are here:

http://www.cs.bham.ac.uk/research/projects/cogaff/talks/#glang

Biosciences May 2010 & SAB2010 Slide 36 Last revised: January 13, 2011

All organisms are information-processors but the information to be processed has changed and so have the means

Types of environment with different information-processing requirements

• Chemical soup
• Soup with detectable gradients
• Soup plus some stable structures (places with good stuff, bad stuff, obstacles, supports, shelters)
• Things that have to be manipulated to be eaten (e.g. disassembled)
• Controllable manipulators
• Things that try to eat you
• Food that tries to escape
• Mates with preferences
• Competitors for food and mates
• Collaborators that need, or can supply, information.

See also http://www.cs.bham.ac.uk/research/projects/cogaff/misc/creativity-boden.html

Biosciences May 2010 & SAB2010 Slide 37 Last revised: January 13, 2011

The role of the environment

Ulric Neisser: “We may have been lavishing too much effort on hypothetical models of the mind and not enough on analyzing the environment that the mind has been shaped to meet.”

Neisser, U. (1976) Cognition and Reality, San Francisco: W. H. Freeman.

Compare: John McCarthy: “The well-designed child” “Evolution solved a different problem than that of starting a baby with no a priori assumptions. ....... “Instead of building babies as Cartesian philosophers taking nothing but their sensations for granted, evolution produced babies with innate prejudices that correspond to facts about the world and babies’ positions in it. Learning starts from these prejudices. What is the world like, and what are these instinctive prejudices?”

http://www-formal.stanford.edu/jmc/child.html Also in AI Journal, December 2008

All biological organisms are solutions to design problems that cannot be specified without specifying in detail the relevant features of the environment.

Turing, surprisingly got this wrong: he thought human-like learning was possible from a “clean slate”.

Biosciences May 2010 & SAB2010 Slide 38 Last revised: January 13, 2011

How to look at the environments of organisms

All biological organisms are solutions to design problems that cannot be specified without specifying in detail the relevant features of the environment.

(This does not imply that there is a designer.)

In order to understand which features of the environment are capable of influencing designs (or more precisely producing “pressures” to alter designs) we have to understand what the problems are that a team of engineers would have to solve – including hardware and software engineers.

Biosciences May 2010 & SAB2010 Slide 39 Last revised: January 13, 2011

That means understanding things like

• What information the organism needs in different parts of the

environment while in different states (hungry, thirsty, escaping, competing, playing, exploring, etc.)

• What forms of representation of that information can be useful for the

purposes of influencing internal and external processes including physical behaviours and information-processing (at the time or in future). in future).

• What information processing mechanisms can make use of the

information

• What sort of architecture can combine a variety of forms of information

processing, some of them running concurrently

(different subsets at different times).

Turing, surprisingly, got this wrong: he thought human-like learning was possible from a “clean slate”. J.J. Gibson understood the general point, but missed many important details. Other authors: Piaget, Fodor, Chomsky, Mandler, Keil, Gopnik, Carey, Tenenbaum, Thomasello, Spelke, Karmiloff-Smith, Pinker, ... (Not all seem to understand how to think about requirements and designs.)

Biosciences May 2010 & SAB2010 Slide 40 Last revised: January 13, 2011

Dynamical systems I

Many researchers who emphasise the importance of embodiment, also emphasise dynamical systems — Especially dynamical systems closely coupled with the environment – Where the nature of the coupling depends on the agent’s morphology and sensory motor systems.

Biosciences May 2010 & SAB2010 Slide 41 Last revised: January 13, 2011

Dynamical systems II

E.g. Perceiving, adaptation, learning, acting, self-monitoring can all involve information processed at multiple levels of abstraction. Hypothetical reasoning: Science, mathematics, philosophy... Some of the more abstract processes may be totally decoupled from the environment – but may produce results that are stored in case they are useful... Which species can do what? – What are intermediate stages: in evolution? in individual development? Do not believe symbol-grounding theory: use theory-tethering instead. (Label proposed by Jackie Chappell).

Biosciences May 2010 & SAB2010 Slide 42 Last revised: January 13, 2011

The 20th C Philosophical breakthrough: Virtual machinery

Brief introduction to the philosophical significance of the technology of virtual machinery (not virtual reality) developed over the last six decades: Processes and events in running virtual machines can be causes and effects, despite being implemented in deterministic physical mechanisms.

Bad model: Good model:

Biosciences May 2010 & SAB2010 Slide 43 Last revised: January 13, 2011

The 20th C Philosophical breakthrough: Virtual machinery

The erroneous picture on the left implies that there is only one-way causation from physical to mental (epiphenomenalism) as in the Mormann & Koch diagram, above.

Bad model: Good model: As the picture on right indicates, we need to think about running virtual machinery that co-exists with, and influences, underlying physical machinery, which it helps to control, even though the virtual machinery is all fully implemented in the physical machinery. I.e. running software can cause changes in physical hardware, just as increases in poverty can cause increases in crimes involving movement of stolen objects.

Biosciences May 2010 & SAB2010 Slide 44 Last revised: January 13, 2011

Springs and levers vs enforced rules

There are some machines in which effects are propagated and constraints are enforced by means of purely physical mechanisms that directly produce the required effects, for instance the system of levers and springs that ensures that

• if a boot lid is moved above a certain height then it is automatically pushed up further

and held there,

• whereas if it is moved down below a certain height then it is pulled further down.

Propagation of effects of changes in a virtual machine and enforcement of constraints on such changes is often very different from physical propagation and constraints, and produced by quite different mechanisms:

• designed to maintain configurations of patterns in switch configurations,
• or to propagate changes in those patterns
• and sometimes to do that with patterns that are not composed directly of combinations
• f switch states,
• but are more abstract structures not directly mapped onto physical switch patterns
• for instance in the layers of structure used in network protocols, or a chess machine

whose rules always ensure that attempts to check its king are blocked in advance.

In such systems the causal mechanisms at work are not visible objects like springs and levers but often rules that are interpreted by hardware or software designed to do whatever the installed rules specify!

Biosciences May 2010 & SAB2010 Slide 45 Last revised: January 13, 2011

Fixed and variable configurations of causes

The collection of constraints and stability patterns in the car boot lid system is fixed: the system remains the same indefinitely

(as long as it doesn’t break or wear out.)

In contrast, collections of rules linking a set of structures, events and processes in a RVM can change frequently, if the system includes a rule interpreter or incremental compiler

contrast programming languages that require each program to be completely specified in advance, then compiled and run (perhaps run many times with different parameters).

For example, the SimAgent toolkit demonstrated in the sheepdog and “emotional” agent demos, allows rules to be edited, added, removed, or reordered while a program is running, and since the conditions and actions of rules can link quite complex structures a system implemented like that may be capable of making extensive changes to itself, which alter the networks of causation that define the nature of the virtual machinery.

Systems built on software mechanism that allow not just data structures, but also active rules and programs to be changed, can grow new architectures, or new VMs. This can happen using

• either mechanisms in which lots of small, blind processes allow complex new things to emerge (as in

patterns of swarming insects or birds)

• or mechanisms with richer semantic competences able to represent and achieve explicit

re-structuring/reprogramming goals – triggered by a sophisticated motivational system.

Biosciences May 2010 & SAB2010 Slide 46 Last revised: January 13, 2011

Granularity differs at different levels

A single indivisible virtual machine process, such as copying a symbol from one abstract memory location to another can involve a large number

• f changes at the electronic level (including hardware error checking):

e.g. electrical pulses along conductors, and switches flipped from one state to another. This is a common feature of mappings from virtual machine structures and events to the physical structures and events that underpin them: typically many distinguishable physical changes correspond to each “minimal” virtual machine change. A more familiar feature is multiple realisability: the same VM process, if repeated, can make use of different physical machine processes on different occasions e.g.

• because the same abstract patterns are mapped onto different parts of the machine at different times,
• because faulty physical components can be replaced by others using different technology, so that a new
• ccurrence of a previous virtual machine event produces a new type of physical process, but interfacing

mechanisms make the differences invisible to other parts of the system. Some philosophers have mistakenly assumed that when a non-physical process “supervenes” on a physical process the two processes are isomorphic: ignoring the differences in granularity described above and

• ther differences – e.g. containment can be circular in VMs but not in physical structures.

One reason why coarser granularity is sometimes essential for human developers and users, is to allow processes to be monitored and managed by human brains that would be unable to think or reason about the low-level details, where far more is happening. For a similar reason, coarse-grained, relatively simple, VMs may be needed in self-monitoring, self-modulating, self-debugging, self-extending machines. (E.g. minds!)

Biosciences May 2010 & SAB2010 Slide 47 Last revised: January 13, 2011

Give some demos of RVMs

E.g.

• toy simulated sheepdog and sheep:

An interactive demo showing a virtual machine with a collection of concurrently active components (sheepdog, sheep) also causally linked to physical devices, e.g. computer mouse and screen (as well as internal registers, memory, etc.).

• toy simulated “emotional agents”:

Another interactive demo where agents interact, change external locations and relationships, change internal states, detect some of their internal states and report them.

Some simple movies are here

http://www.cs.bham.ac.uk/research/projects/poplog/figs/simagent

(including showing non-interactive versions of the above – i.e. video-recordings of the RVMs not the running programs themselves): The demos show that it is possible to generate a running program in which there are various sub-programs driving different entities, e.g. entities inhabiting a 2-D virtual space in which they move, that interact with

• ne another in ways that are continually displayed on a screen, and with which a person can interact by

using mouse or keyboard, directing actions selectively at different entities at different times: e.g. in the sheep-dog program, moving one of the sheep, one of the trees, or the dog, with consequent effects on

• ther things in the virtual machine, because they sense the changes in the moved object (and other objects

that move autonomously) and the perceived changes produce further reactions in the perceiver. This sort of demo is not particularly impressive by current standards but depends on hardware and software developments in the last few decades: it would have been very difficult to produce such a collection of interacting pieces of virtual machinery on computers generally available in 1960s and 1970s.

Biosciences May 2010 & SAB2010 Slide 48 Last revised: January 13, 2011

Some of the relevant technological advances

A small sample of technical developments supporting use of increasingly sophisticated RVMs in computing systems over the last half century:

• The move from bit-level control to control of and by more complex and abstract patterns.
• The move from machine-level instructions to higher level languages (using compilers that ballistically

translate to machine code and especially interpreters that “translate” dynamically, informed by context). A deep difference between compiled and interpreted programs: the compilation process makes the

• riginal program irrelevant, unlike an interpretation process: so altering interpreted program code at

run time can have effects that would not occur if the program had been compiled and run.

• Memory management systems make physical memory reference context-dependent.
• Virtual memory (paging and cacheing) and garbage collection switch virtual memory contents between

faster and slower core memories and backing store, and between different parts of core memory: constantly changing PM/VM mappings. (These support multiple uses of limited resources.)

• Networked file systems change apparent physical locations of files.
• Device interfaces translate physical signals into “standard” RVM signals and vice versa.
• Devices can themselves run virtual machines with buffers, memories, learning capabilities...
• Device drivers (software) handle mappings between higher level and lower level RVMs – and allow

devices to be shared between RVMs (e.g. interfaces to printers, cameras, network devices).

• Context-dependent exception and interrupt handlers distribute causal powers over more functions.
• Non-active processes persist in memory and can have effects on running processes through shared
• structures. (It’s a myth that single-cpu machines cannot support true parallelism.)
• Multi-cpu systems with relocatable RVMs allow VM/PM mappings to be optimised dynamically.
• Multiplicity of concurrent functions continually grows – especially on networked machines.
• Over time, control functions increasingly use monitoring and control of RVM states and processes.

Biosciences May 2010 & SAB2010 Slide 49 Last revised: January 13, 2011

Different requirements for virtual machinery

The different engineering developments supporting new kinds of virtual machinery helped to solve different sorts of problems. E.g.

• Sharing limited physical devices between different users or functions efficiently.
• Optimising allocation of devices of different speeds between sub-tasks.
• Setting interface standards

so that suppliers can produce competing solutions, and new technology can be used for old functions.

• Allowing re-use of design solutions in new contexts.
• Simplifying large scale design tasks by allowing components to “understand” more

complex instructions (telling them what to do, leaving them to work out how to do it).

• Letting abstract functionality be instantiated differently in different contexts

(polymorphism).

• Improving reliability of systems using unreliable components. (E.g. error-checking memory.)
• Allowing information transfer/information sharing to be done without users having to

translate between formats for different devices (especially unix since mid 1970s).

• Simplifying tasks not only for human designers but also for self-monitoring, self-

modulating, self-optimising, self-extending systems and sub-systems.

These are solutions to problems that are inherent in the construction and improvement of complex functioning systems: they are not restricted to artificial systems, or systems built from transistors, or ...

Conjecture: Similar problems were encountered in biological evolution (probably many more problems) and some of the solutions developed were similar, while some were a lot more sophisticated than solutions human engineers have found so far.

Biosciences May 2010 & SAB2010 Slide 50 Last revised: January 13, 2011

Benefits of using running VMs

We don’t know exactly what problems evolution faced, what solutions it came up with, and what mechanisms it used, in creating virtual machinery running in animals, to control increasingly complex biological organisms (and societies). Perhaps we can learn from the problems human engineers faced, and the solutions they found.

• For example, a chess-playing computer can look for moves that achieve a win, or

make a win likely.

• In performing that search it need not have the ability to represent the physical details
• f either the end result of such a search process or the physical details of the process
• f searching.
• Likewise a machine that aims to observe and understand processes running in itself is

likely to do better by using a level of abstraction that has all the details needed for self monitoring, without representing all the physical details of the process.

• It turns out to be essential for designers of self-monitoring intelligent systems not to

make any assumptions in advance about how they work, but to focus only on what they do.

Biosciences May 2010 & SAB2010 Slide 51 Last revised: January 13, 2011

Virtual machinery and causation

Virtual machinery works because “high level” events in a RVM can control both other virtual machinery and physical machinery.

Accordingly, buggy (wrongly designed) interactions between bits of virtual machinery can lead to disasters, even though nothing is wrong with the hardware. As stated previously Processes and events in running virtual machines can be causes and effects, despite being implemented in deterministic physical mechanisms.

Engineers (but not most philosophers, psychologists, neuroscientists?) now understand how running virtual machinery can co-exist with, and influence, underlying physical machinery, which it helps to control, even though the virtual machinery is all fully implemented in the physical machinery.

System designers who are concerned with providing the functionality, making it robust, or modifying it to cope with changing circumstances, often ignore the hardware involved.

They think about, and write code to deal with, events like: information arriving, a context changing, a request being received, a plan being executed, a rule being followed with unexpected results, a rule being violated, etc. In doing that they depend on vast amounts of technology some of which they may not even be aware of, e.g. the use of cache mechanisms, garbage collectors, interrupt handlers, device drivers, etc.

Biosciences May 2010 & SAB2010 Slide 52 Last revised: January 13, 2011

Two major kinds of running VM

There are various types of VM with different capabilities and different roles in larger systems. Two important classes are: (1) Specific function/application VMs

E.g. a chess-playing VM, a word-processor, an email front-end.

(2) Platform VMs

Capable of supporting a variety of other VMs Some capable of extending/modifying themselves. E.g. Operating systems (e.g. Linux, Unix, OSX, Windows, VMS, ...) Language VMs (e.g. Lisp VM. Prolog VM, Java VM, Pop-11 VM)

NB: we are talking about running instances, not abstract specifications.

It seems that there are some biological platform VMs, which get extended in various ways. An important research problem is to investigate the various types, their functions, which species use them, how they evolved, how they develop in individuals, etc. One type of development of biological VM capability involves learning new forms of self-monitoring; another involves new languages and notations.

E.g. learning a new programming language.

Biosciences May 2010 & SAB2010 Slide 53 Last revised: January 13, 2011

Natural and Artificial Platform VMs

• Platform VMs designed by human engineers provide a basis for constructing new VMs

that are implemented in terms of the facilities provided by the platform VM: But most such extensions do not arise spontaneously.

The operating system on your PC can be left running for many years and over time you and others may design and install different software packages, or attach new hardware devices along with device drivers for them. If left to itself, a normal operating system will just go on forever waiting for instructions, without initiating any major extensions, though some of them are designed to detect the availability of new versions of old subsystems and download and install them.

• Biological platform VMs, however, are not extended by external designers:

They have to build and extend themselves

(partly on the basis of external influences from both the physical environment and conspecifics). The requirements to support this have never, as far as I know, been identified. The problem is not addressed by research in developmental psychology on which concepts, knowledge or competences are innate. Some of the requirements for a “well-designed child” were discussed by McCarthy in this paper http://www-formal.stanford.edu/jmc/child.html (written 1996 and published in 2008).

• In humans, biological platform VMs seem to grow throughout infancy and childhood,

and for some people (e.g. academics) continue being extended until late in life.

The extensions support new competences of many kinds, including manipulative and perceptual competences, linguistic, musical and artistic competences, mathematical competences, extended ontologies, new planning and reasoning capabilities, new forms of motivation, new control regimes,

Biosciences May 2010 & SAB2010 Slide 54 Last revised: January 13, 2011

Biological VMs have to grow themselves

Humans and some other species require layered construction of competences and meta-competences in a layered architecture, e.g. starting with left-most routes to action, then adding more, as shown. Not core competences as normally construed: core architecture-building competences, metacompetences, ... Work done with Jackie Chappell (IJUC, 2007) – Chris Miall helped with diagram. “Natural and artificial meta-configured altricial information-processing systems” http://www.cs.bham.ac.uk/research/projects/cosy/papers/#tr0609,

Biosciences May 2010 & SAB2010 Slide 55 Last revised: January 13, 2011

Reactive vs deliberative interacting patterns

A Conway ”life” machine uses real or simulated concurrency: behaviour of each square depends only on the previous states of its eight neighbours and nothing else.

On a computer the concurrency is achieved by time-sharing, but it is still real concurrency. Consider what happens when two virtual machines running on a computer compete in a chess game, sharing a virtual chess board, and interacting through moves on the board, each can sense or alter the state of any part of the (simulated) chess board.

• In general, programs on a computer are not restricted to local interactions.
• In some cases, the interacting processes are purely reactive: on every cycle every

square immediately reacts to the previous pattern formed by its neighbours.

• If two instances of a chess program (or instances of different chess programs) interact

by playing chess in the same computer, their behaviour is typically no longer purely

• reactive. Good ones will often have to search among possible sequences of future

moves to find a good next move – and only then actually move.

In addition, one or both of the chess virtual machines may do some searching in advance while waiting for the opponent’s next move.

• Then each instance is a VM with its own internal states and processes interacting

richly, and a less rich interaction with the other VM is mediated by changes in the shared board state (represented by an abstract data-structure).

For more on varieties of deliberation see: http://www.cs.bham.ac.uk/research/projects/cosy/papers/#dp0604

Biosciences May 2010 & SAB2010 Slide 56 Last revised: January 13, 2011

Intentionality in a virtual machine

A running chess program (a VM) takes in information about the state of the board after the opponent moves, and builds or modifies internal structures that it uses to represent the board and the chess pieces on it, and their relationships, including threats, opportunities, possible traps, etc.

• In particular it uses those representations in attempting to achieve its goals.

So, unlike the interacting Conway patterns mentioned earlier, some of the patterns in the chess virtual machine are treated by the machine as representations, that refer to something.

• During deliberation, some created patterns will be treated as referring to non-existent

but possible future board states, and as options for moves in those states.

They are treated that way insofar as they are used in considering and evaluating possible future move sequences in order to choose a move which will either avoid defeat (if there is a threat) or which has a chance of leading to victory (check-mate against the opponent).

• In this case the chess VM, unlike the simplest interacting Conway patterns, exhibits

intentionality: the ability to refer. (NB. The programmer need not know about the details.)

Since the Conway mechanism is capable of implementing arbitrary Turing machines, it could in principle implement two interacting chess virtual machines, so there could be intentionality in virtual machines running on a Conway machine – probably requiring a very big fairly slow machine.

• The intentionality of chess VMs is relatively simple because they have relatively few

types of goal, relatively few preferences, and their options for perceiving and acting are limited by being constrained to play chess:

For a human-like, or chimp-like, robot the possibilities would be much richer, and a far more complex architecture would be required. See

http://www.cs.bham.ac.uk/research/projects/cogaff/03.html#200307

Biosciences May 2010 & SAB2010 Slide 57 Last revised: January 13, 2011

Adding meta-semantic competences

If a virtual machine playing chess not only thinks about possible board states and possible moves, and winning, moving, threats, traps, etc. but also thinks about what the opponent might be thinking, then that requires meta-semantic competences: the ability to represent things that themselves represent and use information.

• It is very likely that biological evolution produced meta-semantic competences in some
• rganisms other than humans because treating other organisms (prey, predators,

conspecifics to collaborate with, and offspring as they learn) as mere physical systems, ignoring their information-processing capabilities, will not work well (e.g. hunting intelligent prey, or avoiding intelligent predators).

• Another application for meta-semantic competences is self-monitoring, self evaluation,

self-criticism, self-debugging: you can’t detect and remedy flaws in your thinking, reasoning, planning, hypotheses etc. if you are not able to represent yourself as an information user.

• It is often assumed that social meta-semantic competences must have evolved first,

but that’s just an assumption: it is arguable that self-monitoring meta-semantic competences must have evolved first

e.g. because an individual has relatively direct access to (some of) its own information-processing whereas the specifics of processing in others has to be inferred in a very indirect way (even if evolution produced the tendency to use information about others using information).

See A. Sloman, 1979, The primacy of non-communicative language, http://www.cs.bham.ac.uk/research/projects/cogaff/81-95.html#43

Biosciences May 2010 & SAB2010 Slide 58 Last revised: January 13, 2011

BIOLOGICAL CONJECTURE: Over time, control functions increasingly used monitoring and control of VM states and processes,

in addition to monitoring and control of physical states and processes. CONJECTURE:

• The pressures for such developments that drive human engineers towards more and

more “virtual” machinery of different sorts, were just as strong, in biological evolution:

• as biological machines and their control functions became more and more complex
• with increasingly complex decisions taken “at run time”
• about how to process sensory information, what ontologies to use, what information to

store, how to use the information, how to generate hypotheses, goals, plans, etc.

• how to use them
• how to detect bugs in them, and debug them, ...

Controlling very large numbers of interacting tiny physical subsystems (e.g. transistors, or neurons) directly is far too difficult. But the solution involving RVMs depends crucially on finding the right, re-usable, levels of abstraction, and modules. Whether there are any, and what they are depends on the type of environment.

Biosciences May 2010 & SAB2010 Slide 59 Last revised: January 13, 2011

Work to be done:

Biology, psychology, neuroscience, robotics

There is much work still to be done. That includes finding out precisely what the problems were that evolution solved and how they are solved in organisms, and why future intelligent robots will need similar solutions.

There are more slide presentations on related topics here:

http://www.cs.bham.ac.uk/research/projects/cogaff/talks/

Many of the papers in the Birmingham CogAff project (Cognition and Affect) are relevant, especially papers on architectures.

http://www.cs.bham.ac.uk/research/projects/cogaff/

But the problem of explaining how a genome can specify types of virtual machinery to be developed in individuals, including types that are partly determined by the environment at various stages of development is very difficult. We need to understand much more about the evolution and development of virtual machinery.

See Jackie Chappell and Aaron Sloman, “Natural and artificial meta-configured altricial information-processing systems” http://www.cs.bham.ac.uk/research/projects/cosy/papers/#tr0609 (IJUC, 2007)

Biosciences May 2010 & SAB2010 Slide 60 Last revised: January 13, 2011

To gain deep understanding, don’t study just one species

Trying to do AI as science by developing only one type of system

(e.g. “human level AI”)

is like trying to do physics by investigating only how things behave near the leaning tower of Pisa. Or studying only the motion of planets.

we need to understand spaces of possibilities and the tradeoffs between alternative designs: so look at different species. Don’t assume all effective information-processing mechanisms have to be rational

(as in Dennett’s “intentional stance”, Newell’s “knowledge level”.)

Engineers need to build reflexes into complex machinery to cope with the unexpected. Likewise, evolution provided useful reflexes: reflexes are neither rational nor irrational. Some useful reflexes are cognitive. Some even meta-cognitive. Including reflexes that generate goals/motives warnings, things to remember, etc. Not all goals are chosen because the individual knows associated rewards/costs/benefits:

See

http://www.cs.bham.ac.uk/research/projects/cogaff/misc/architecture-based-motivation.html

Architecture-based motivation vs reward-based motivation.

Biosciences May 2010 & SAB2010 Slide 61 Last revised: January 13, 2011

What is a machine (natural or artificial)? (1)

The word “machine” often refers to a complex enduring entity with parts

(possibly a changing set of parts)

that interact causally[*] with other parts, and other “external” things, as they change their properties and relationships.

[*]Causation is discussed later.

The internal and external interactions may be

• discrete or continuous,
• concurrent (most machines), or sequential (e.g. row of dominoes, a fuse(?))
• if concurrent then synchronised or asynchronous

In Turing machines, everything is:

• Internal
• Discrete
• Sequential
• Synchronous

Concurrent and synchronized TMs are equivalent to sequential TMs.

I.e. parallelism in TMs adds nothing new.

But some machines have concurrent parts that are not synchronised, so they are not TMs, even if they have TM-like components.

And systems interacting with a physical or social environment are not TMs, since a TM, by definition, is a self-contained: machine table+tape.

Biosciences May 2010 & SAB2010 Slide 62 Last revised: January 13, 2011

What is a machine (natural or artificial)? (2)

The word “machine” often refers to a complex enduring entity with parts

(possibly a changing set of parts)

that interact causally with other parts, and other “external” things, as they change their properties and relationships.

The internal and external interactions may be

• discrete or continuous,
• concurrent (most machines), or sequential (e.g. row of dominoes, a fuse(?))
• if concurrent then synchronised or asynchronous

NOTES

• 1. Machines, in this general sense, do not have to be artificial, or man-made, or

deliberately designed to do what they do.

• 2. The perception of machines and how they work is one of the important functions of

human visual perception, and haptic/tactile perception, (possibly also in some other species).

That includes the perception of structures, processes and causal relationships (proto-affordances). This is generally ignored by vision researchers. Perception of affordances is a special case of this. E.g. See Architectural and Representational Requirements for Seeing Processes, Proto-affordances and Affordances, http://drops.dagstuhl.de/opus/volltexte/2008/1656

Biosciences May 2010 & SAB2010 Slide 63 Last revised: January 13, 2011

Typical features of machines (natural and artificial):

Machines

• can have various degrees and kinds of complexity

(often hierarchical – machines composed of machines)

• allow changes/processes to occur within them

usually concurrent changes (e.g. gear wheels turning, ends of lever moving in opposite directions)

• can acquire, manipulate, use, produce, and/or transfer matter, energy or information.
• include processes that involve not mere change, but also causation

– within the machine E.g. parts moving other parts, forces transmitted, information stored, retrieved, derived

• r transmitted, parts controlling or activating, other parts.

– partly within the environment E.g. if there are sensors, motors, and communication channels – involving matter, motion, forces, energy, information, ... and more

• are usually embedded in a complex environment with which they interact. Often the

boundary between machine and environment is different for different sub-systems of the machine. As every mathematician knows, you can use pen and paper as an extension of your mind.

Sloman IJCAI 1971: http://www.cs.bham.ac.uk/research/cogaff/04.html#200407

• may include some internal processes whose effects are not externally detectable,

e.g. a machine playing chess with itself and learning to play better as a result. In some cases the unobservable internal processes can be inferred indirectly. (More on this later)

Biosciences May 2010 & SAB2010 Slide 64 Last revised: January 13, 2011

What is a physical machine (PM)?

Some, but not all, machines satisfying the previous definition are physical. If a machine and its operations (processes, and causal relationships) are fully describable using concepts of the physical sciences (plus mathematics), it is a physical machine (PM).

That’s a first draft specification. I’ll contrast that with a kind of machine whose parts and processes are not fully describable within the language of the physical sciences: some additional concepts are required. E.g. concepts like “winning” in a game or “correct spelling” in a document, or “attempting” to achieve something, which, I suggest, cannot be defined in the language of the physical sciences. (There is probably a variant definition that does not mention concepts, but I am not sure.)

The contents of the physical sciences, expand over time, so the broadest notion of “physical machine” must refer to the indefinite future. Examples of physical machines include: levers, assemblages of gears, mechanical clocks, audio amplifiers, electronic devices, wireless control systems, clouds, tornadoes, plate tectonic systems, atoms, bacteria, brains, and myriad molecular machines in living organisms. There is much we don’t know about what sorts of machine can be built out of chemical

• components. E.g. read recent issues of Scientific American.

Biosciences May 2010 & SAB2010 Slide 65 Last revised: January 13, 2011

Virtuality: absolute and relative

I shall first introduce a distinction between machines that are physical and machines that are not, even though the ones that are not are “fully implemented” in physical machines. Later we’ll see that that is one of many examples of “layering” – one aspect of reality is layered on another, so that certain things are virtual relative to others. But first we start with a single distinction.

Our preliminary notion of a running virtual machine (RVM), refined later, is defined as a machine in the sense defined earlier, but is not a physical machine, in the sense of “physical machine” defined above: i.e. a RVM can be complex, with parts that interact with one another and with things

• utside the machine, but describing the parts and their operations requires use of

concepts that are not definable in terms of concepts of the physical sciences.

E.g. spelling checker, chess program, proof checker, winning, invalid inference.

This notion will now be elaborated. It is relative to a concept of a physical science, a concept that has changed over centuries, making the distinction a fluid one.

Biosciences May 2010 & SAB2010 Slide 66 Last revised: January 13, 2011

Non-physically-definable (NPD) concepts

Certain states, processes and interactions of some machines can be described fully only if we use concepts that are not definable in terms of concepts

• f the physical sciences

(E.g. not definable in terms of the concepts of physics, chemistry, plus mathematics.)

Information-processing machines are examples.

“Information” is not used here in Shannon’s sense, but in the sense that includes “reference”, “meaning”, with properties and relations like: truth, consistency, implication, contradiction, These concepts are not definable in terms of concepts of the physical sciences.

Though every information-using machine must be implemented (realised) in a physical machine.

See http://www.cs.bham.ac.uk/research/projects/cogaff/misc/whats-information.html Non-physical machines include: socio-economic machines, ecosystems, many biological control systems, and many of the things that run in computers, including games, spelling checkers, email systems, time-management systems, operating systems and networking systems.

Biosciences May 2010 & SAB2010 Slide 67 Last revised: January 13, 2011

More on non-physically describable machines

Non-Physically-Describable Machines (NPDMs) are the subject matter

• f common sense, gossip, novels, plays, legends, history, the social

sciences and economics, psychology, and various aspects of biology.

An important common feature is use of non-physically-definable concepts. Examples of such non-physically-definable concepts:

“information”, “inference”, “contradiction”, “strategy”, “desire”, “belief”, “mood”, “promise”, “contract”, “checking spelling”, “file access violation”, “sending email”, “playing chess”, “winning”, “threat”, “defence”, “plan”, “poverty”, “crime”, “economic recession”, “election”, “war”, ... For now I’ll take that indefinability as obvious: It would take too long to explain and defend. This is connected with the falsity of “concept empiricism” and “symbol grounding theory”. See

http://www.cs.bham.ac.uk/research/projects/cogaff/talks/#models http://www.cs.bham.ac.uk/research/projects/cogaff/misc/whats-information.html

In computer science and software engineering NPDMs are often called “virtual machines” (terminology possibly derived from some of the earliest examples: virtual memory systems). This terminology is unfortunate – since the word “virtual” can suggest that such machines do not really exist – like the entities represented in virtual reality systems. Nevertheless we are stuck with the word. (Like the misleading phrase “Artificial Intelligence”, which labels a field that includes the study and modelling of natural intelligence.) Later we’ll talk about layers of relative virtuality.

Biosciences May 2010 & SAB2010 Slide 68 Last revised: January 13, 2011

Two sorts of interaction between patterns

• We can impose patterns on things we see: e.g. rows of dots, stars and planets,

moving clouds, mixtures of shadows and sunlight on a forest floor – where some pattern events may appear to be causing other pattern events.

• We may fancy that one piece of shadow chases another, or that one bounces off

another, or that an arrow shaped shadow is pointing at a patch of light, whereas in fact there is no chasing or bouncing or pointing

• But the patterns of light and shade do not interact: they are by-products of interactions

between portions of the tree, portions of the air blowing through or against them, and light shining down onto or between them.

• In particular, there is nothing producing the portion of shadow on the basis of

information about where the patch of light is.

• One way to think of what has been happening in the world of computing

systems is that

(a) we have been learning more and more about how to make changing abstract patterns that really do influence other patterns, (b) in many cases what those patterns are, and what they do, cannot be described using only the language of the physical sciences, e.g. if they are: – trying to win a game, – formulating questions, – making plans, – referring to other patterns, or even to themselves (c) Those patterns can also influence physical machines on which they “run”.

Biosciences May 2010 & SAB2010 Slide 69 Last revised: January 13, 2011

An old, over-simple, idea.

The idea of the analogy expressed in this diagram is very old, but we are only slowly understanding the variety of phenomena on the left hand side, extending

• ur appreciation of what might be going on on the right.

The simple-minded notion that the left hand side involves a program in a computer is seriously deficient, (a) because it ignores the requirement for the program to be running and (b) because we know now that there are far more types of information-processing system than a computer running a single program.

Simple-minded views about virtual machines lead to easily refutable computational theories of mind, e.g. the theory that virtual machines are simple finite state machines, as illustrated on the right (“Atomic state functionalism”). See

Ned Block: Functionalism http://cogprints.org/235/

• A. Sloman, The mind as a control system, 1993,

http://www.cs.bham.ac.uk/research/projects/cogaff/81-95.html#18

Forget what you have learnt about Turing machines: that’s a simple abstraction which is surprisingly useful for theorising about classes

• f computations – but not so useful for modelling complex

multi-component systems interacting asynchronously with a rich and complex environment.

Biosciences May 2010 & SAB2010 Slide 70 Last revised: January 13, 2011

More realistic models

As crudely indicated here we need to allow multiple concurrent inputs (blue) and outputs (red), multiple interacting subsystems, some discrete some continuous, with the ability to spawn new subsystems/processes as needed. Artificial VM on artificial PM Biological VM on biological PM

In both cases there are multiple feedback loops involving the environment.

Biosciences May 2010 & SAB2010 Slide 71 Last revised: January 13, 2011

Warning from biology:

Don’t expect a sharp divide between systems using only physical machines and those also using virtual machines: biology provides intermediate cases for most distinctions,

e.g. is a homeostatic control loop a VM?

Neither biology nor engineering needs to respect philosophers’ desires for simple classification schemes:

there tend to be many small discontinuities rather than just a few big ones. But differences across multiple steps can be huge.

Biosciences May 2010 & SAB2010 Slide 72 Last revised: January 13, 2011

RVMs with temporarily or partly ‘decoupled’ components

A challenge for philosophy of science and psychological methodology.

• “Decoupled” subsystems may exist and process information, even though they have

no connection with sensors or motors.

• Theories referring to them cannot be decisively proved or refuted.

Compare Lakatos on methodology of scientific research programmes

• For instance, a machine playing games of chess with itself, or investigating

mathematical theorems, e.g. in number theory.

• Some complex systems “express” some of what is going on in their VM states and

processes through externally visible behaviours.

However, it is also possible for internal VM processes to have a richness that cannot be expressed externally using the available bandwidth for effectors. Likewise sensor data may merely introduce minor perturbations in what is a rich and complex

• ngoing internal process.

This transforms the requirements for rational discussion of some old philosophical problems about the relationship between mind and body:

E.g. some mental processes need have no behavioural manifestations, though they might, in principle, be detected using ‘decompiling’ techniques with non-invasive internal physical monitoring. (This may be impossible in practice, or at best only a matter of partly testable conjecture. Compare theoretical physics.)

Biosciences May 2010 & SAB2010 Slide 73 Last revised: January 13, 2011

How does all that work?

• We have learnt how to set up physical mechanisms that enforce constraints between

abstract process patterns (unlike mechanisms that merely enforce constraints between physical or geometric relations).

• Chains of such constraints can have complex indirect effects linking different

process-patterns.

• Some interactions involve not only causation but also meaning: patterns are

interpreted by processes in the machine as including descriptive information (e.g. testable conditions) and control information (e.g. specifying what to do). See “What enables a machine to understand?” (IJCAI 1985)

http://www.cs.bham.ac.uk/research/projects/cogaff/81-95.html#4

Could biological evolution have solved similar problems?

Biosciences May 2010 & SAB2010 Slide 74 Last revised: January 13, 2011

Higher ⇐

⇒ lower virtual interfaces at different levels

Starting with simple physical devices implementing interacting discrete patterns, we have built layers of interacting patterns of ever increasing spatial and temporal complexity, with more and more varied functionality.

• Physical devices can constrain continuously varying states so as to allow only a small

number of discrete stable states (e.g. only two)

(e.g. using mechanical ratchets, electronic valves (tubes), aligned magnetic molecules, transistors etc.)

• Networks of such devices can constrain relationships between discrete patterns.

E.g. (the ABCD/XY example) a constraint can ensure that if devices A and B are in states X and Y respectively then devices C and D will be in states Y and X (with or without other constraints). So, a device network can rule out some physically possible combinations of states of components, and a new pattern in part of the network will cause pattern-changes elsewhere via the constraints. Compare: one end of a rigid lever moving down or up causes the other end to be moving up or down.

• Such networks can form dynamical systems with limited possible trajectories,

constraining both the possible patterns and the possible sequences of patterns.

• A network of internal devices can link external interfaces (input and output devices)

thereby limiting the relationships that can exist between patterns of inputs and patterns of outputs, and also limiting possible sequences of input-output patterns.

• Patterns in one part of the system can have meaning for another part, e.g.

– constraining behaviour (e.g. where the pattern expresses a program or ruleset) or – describing something (e.g. where the pattern represents a testable condition)

• Such patterns and uses of such patterns in interacting computing systems may result

from design (e.g. programming) or from self-organising (learning, evolving) systems.

• Some useful patterns need not be describable in the language of physics.

Biosciences May 2010 & SAB2010 Slide 75 Last revised: January 13, 2011

What follows from all this?

• There are many easily checked empirical facts about human experience (e.g. cube

and duck-rabbit ambiguities) that support claims about the existence of introspectively accessible entities, often described as privately accessible contents of consciousness.

Various labels are used for these entities: “phenomenal consciousness”, “qualia” (singular “quale”), “sense-data”, “sensibilia”, “what it is like to be/feel X” (and others). For a useful, but partial, overview see http://en.wikipedia.org/wiki/Qualia

• What is not clear is what exactly follows from the empirical facts, and how best they

can be described and explained.

• Earlier slides demonstrated some of the empirical facts about contents of

consciousness (e.g. necker flips) that raise a scientific problem of explaining how such entities arise and how they are related to non-mental mechanisms, e.g. brains.

• Philosophers and scientists, understandably (in the past) ignorant of what we now

know about virtual machinery have referred to “the explanatory gap” later re-described by Chalmers as “the hard problem” of consciousness: a problem for Darwin

• Unfortunately, some philosophers, trying to characterise what needs to be explained

have ended up discussing something incoherent.

E.g. qualia, or phenomenal consciousness defined as incapable of causal/functional relations.

For more on this see my online presentations: http://www.cs.bham.ac.uk/research/projects/cogaff/talks/

Biosciences May 2010 & SAB2010 Slide 76 Last revised: January 13, 2011

Confusions about infallibility

There is a widely believed myth that consciousness involves a kind of infallibility, which is part of what defines the problem of consciousness.

There are two interpretations of the infallibility claim.

• On one interpretation it is a trivial tautology

“We are infallible about what we experience.” “We have ‘direct access’ to our own states of consciousness so we cannot be mistaken about them.” Descartes: “I can doubt everything but not that I am doubting.” “I can be wrong about whether there is a dagger before me, but not about whether there seems to me to be a dagger before me.” But this is exactly like a certain sort of infallibility of every measuring device – it cannot be mistaken about what it reports! A voltmeter can be wrong about what the voltage is, if it is faulty, but it cannot be wrong about what it reports the voltage to be. Likewise, we can be mistaken about what we have seen, but not about what we seem to have seen. However, nothing of any interest follows from this tautology. In particular, it does not rule out being mistaken about what we have perceived.

• On the other interpretation it is an empirical claim.

But the claim is false as shown by this experiment (which does not work for everyone):

http://www.cs.bham.ac.uk/research/projects/cogaff/misc/unconscious-seeing.html

and by blindsight phenomena, e.g.

http://www.boston.com/bostonglobe/ideas/brainiac/2010/05/blindsight.html

This shows we need to distinguish what we are conscious of and what we are conscious of being conscious of. I suspect the latter evolved much later and is missing in most conscious organisms.

Biosciences May 2010 & SAB2010 Slide 77 Last revised: January 13, 2011

Emerging varieties of functionality

Computer scientists and engineers and AI/Robotics researchers have been learning to add more and more kinds of control, kinds of pattern, and ways of interpreting patterns of varying levels of abstraction.

• A simple machine may repeatedly take in some pattern and output a derived pattern,

e.g. computing the solution to an arithmetical problem.

• More complex machines can take in a pattern and a derivation-specification (program)

and output a derived pattern that depends on both.

• Other machines can continually receive inputs (e.g. from digitised sensors) and

continually generate outputs (e.g. to digitally controlled motors).

• More sophisticated machines can

– solve new problems by searching for new ways of relating inputs to outputs, i.e. learning; – interpret some patterns as referring to the contents of the machine (using a somatic ontology) and

• thers to independently existing external entities, events, processes (using an exosomatic ontology)

– extend their ontologies and theories about the nature and interactions of external entities – perform tasks in parallel, coordinating them, – monitor and control some of their own operations – even interrupting, modulating, aborting, etc. (Including introspecting some of their sensory and other information contents: qualia.) – develop meta-semantic ontologies for representing and reasoning about thinking, planning, learning, communicating, motives, preferences, ... – acquire their own goals and preferences, extending self-modulation, autonomy, unpredictability, ... – develop new architectures which combine multiple concurrently active subsystems. – form societies, coalitions, partnerships ... etc.

• Biological evolution did all this and more, long before we started learning how to do it.

Biosciences May 2010 & SAB2010 Slide 78 Last revised: January 13, 2011

Causal networks linking layered patterns

How can events in virtual machines be causes as well as effects, even causing physical changes?

The answer is through use of mechanisms that allow distinct patterns of states and sequences

• f patterns to be linked via strong constraints to other patterns of states and

sequences of patterns (as in the ABCD/XY example, and the Conway machines, mentioned above). (Some VMs may use probabilistic/stochastic constraints.) What many people find hard to believe is that this can work for a virtual machine whose internal architecture allows for divisions of functionality corresponding to a host of functional divisions familiar in human minds, including

• interpreting physical structures or abstract patterns as referring to something (intentionality)
• generation of motives,
• selection of motives,
• adoption of plans or actions,
• perceiving things in the environment,
• introspecting perceptual structures and their changes,
• extending ontologies,
• forming generalisations,
• developing explanatory theories,
• making inferences,
• formulating questions,
• and many more.

Biosciences May 2010 & SAB2010 Slide 79 Last revised: January 13, 2011

Biological unknowns: Research needed

Many people now take it for granted that organisms are information-processing systems, but much is still not known, e.g. about the varieties of low level machinery available (at molecular and neuronal mechanisms) and the patterns of organisation for purposes of acquiring and using information and controlling internal functions and external behaviours.

Steve Burbeck’s web site raises many of the issues: “All living organisms, from single cells in pond water to humans, survive by constantly processing information about threats and opportunities in the world around them. For example, single-cell E-coli bacteria have a sophisticated chemical sensor patch on one end that processes several different aspects of its environment and biases its movement toward attractant and away from repellent

• chemicals. At a cellular level, the information processing machinery of life is a complex network of

thousands of genes and gene-expression control pathways that dynamically adapt the cell’s function to its environment.”

http://evolutionofcomputing.org/Multicellular/BiologicalInformationProcessing.html

“Nature offers many familiar examples of emergence, and the Internet is creating more. The following examples of emergent systems in nature illustrate the kinds of feedback between individual elements of natural systems that give rise to surprising ordered behavior. They also illustrate the trade off between the number of elements involved in the emergent system and the complexity of their individual interactions. The more complex the interactions between elements, the fewer elements are needed for a higher-level phenomenon to emerge. ... networks of computers support many sorts of emergent meta-level behavior because computers interact in far more complex ways than air and water molecules or particles of sand ... Some of this emergent behavior is desirable and/or intentional, and some (bugs, computer viruses, dangerous botnets, and cyber-warfare) are not.”

http://evolutionofcomputing.org/Multicellular/Emergence.html

Biosciences May 2010 & SAB2010 Slide 80 Last revised: January 13, 2011

Closing Huxley’s Explanatory Gap

If we can learn more about:

• varieties of virtual machinery and their roles in generating, controlling, modulating and

extending behaviours in organisms;

• how they are implemented in various types of biological organism;
• how their features can be specified in a genome (e.g. the control mechanisms for

mating, web-making, and eating in a spider seem, for many species to be genetically determined, although specific behaviours are adapted to the precise details of environment);

• how in some species the virtual machinery instead of being fully specified genetically

is built up within an individual as a result of operating of genetic, environmental and cultural processes (see Chappell and Sloman, IJUC, 2007, mentioned above);

• how and why self-monitoring mechanisms came to include mechanisms able to focus
• n intermediate information-structures within sensory/perceptual sub-systems

(e.g. how things look, how they feel, how they sound, etc.)

then we may be able to understand how a Darwinian evolutionary process that is already demonstrably able to explain much of the evolution of physical form might be extended to explain evolution of information processing capabilities, including the phenomena that lead to philosophical theories of consciousness. But we should not expect there to be one thing, one it that evolved. Darwin and his contemporaries knew nothing about virtual machines, alas.

Biosciences May 2010 & SAB2010 Slide 81 Last revised: January 13, 2011

Importance and implications of VMs

There are additional slides available on Virtual Machines, e.g.

http://www.cs.bham.ac.uk/research/projects/cogaff/talks/#mos09 Virtual Machines and the Metaphysics of Science Expanded version of presentation at: Metaphysics of Science’09)

Topics include:

• Explanation of the importance of virtual machines in sophisticated control systems

with self-monitoring and self-modulating capabilities.

• Why such machines need something like “access consciousness”/qualia – and why

they too generate an explanatory gap – a gap bridged by a lot of sophisticated hardware and software engineering developed over a long time.

• In such machines, the explanations that we already have are much deeper than mere

correlations: we know how the physical and virtual machinery are related, and what difference would be made by different designs.

Biosciences May 2010 & SAB2010 Slide 82 Last revised: January 13, 2011

More to read

We need much better understanding of nature-nurture issues, and requirements for educational systems. John McCarthy on “The well-designed child”.

http://www-formal.stanford.edu/jmc/child.html

Chappell and Sloman on “Natural and artificial meta-configured altricial information-processing systems”

http://www.cs.bham.ac.uk/research/projects/cosy/papers/#tr0609

Expand on varieties of metacognition, and differences between introspection and other aspects of metacognition.

http://www.cs.bham.ac.uk/research/projects/cosy/papers/#tr0803

See other presentations in

http://www.cs.bham.ac.uk/research/projects/cogaff/talks/

and CogAff and CoSy papers:

http://www.cs.bham.ac.uk/research/projects/cogaff/ http://www.cs.bham.ac.uk/research/projects/cosy/papers/

Further Reading Novelists have traditionally regarded themselves as the experts on consciousness, with some

• justification. See for example, David Lodge’s essays and his novel on consciousness:

David Lodge, Consciousness and the Novel: Connected Essays, Secker & Warburg, London, 2002. David Lodge, Thinks ...., Penguin Books, 2002. A huge and important topic: disorders of consciousness, self-consciousness and control. We need to explain development of kind of self-awareness that enables people to tell the difference between what they treat as empirical generalisations and what they understand as (mathematically) provable – e.g. facts about topological relations, geometry, mechanics, and numbers. (The roots of mathematical thinking.)

Biosciences May 2010 & SAB2010 Slide 83 Last revised: January 13, 2011

Summary

Over the last six or seven decades there have been a lot of separate developments adding functionality of different sorts to computing systems including (in no significant order):

memory management, paging, cacheing, interfaces of many kinds, interfacing protocols, device drivers, adaptive schedulers, privilege mechanisms, resource control mechanisms, file-management systems, interpreters, compilers and run-time systems for a wide variety of types of programming language, garbage collectors, varied types of data-structure and operations on them, tracing and debugging tools, pipes, sockets, shared memory systems, firewalls, virus checkers, security systems, network protocols,

• perating systems, application development systems, etc. etc.

All this is very familiar to computer scientists and software engineers, though different experts know about different sub-sets of these developments and the whole package is not often described adequately. In a way it is very familiar to millions of users, who are incapable of describing what they are using. A consequence of all these developments is that we can now have, in addition to all the physical computing machinery that we use, varying collections of non-physical machinery made up of various kinds of interacting components with causal powers that operate in parallel with the causal powers of the underlying machines, and can help to control those physical machines, but with different kinds of granularity and different kinds of functionality from the physical machines. The mathematical abstractions (Abstract Virtual Machines – AVMs) that are sometimes called virtual machines (e.g. a universal turing machine, the java virtual machine, the/a linux virtual machine, the Prolog virtual machine) are very different from the running virtual machines (RVMs) that actually do things. Each AVM can typically have many instances that are RVMs, doing different things because the instances can differ in significant ways.

Biosciences May 2010 & SAB2010 Slide 84 Last revised: January 13, 2011

References

Maynard Smith, J., & Szathm´ ary, E. (1995). The Major Transitions in Evolution. Oxford, England:: Oxford University Press.